A Comprehensive Pipeline for Reliable Multi-Animal Position Tracking During Mouse Social Interactions Restricted; Files Only
Feng, Jifan (Spring 2024)
Published
Abstract
Social behaviors in mice are inherently complex and high-dimensional with a myriad of intricate and unique behaviors. Whether our application is to relate interactions to neural data or parcellate behavioral signatures, there is a need for efficient analysis of videos that capture the interactions between the animals. These goals are often restricted by our ability to painstakingly mark videos that capture such interactions. Manual annotation of behavior is prone to human errors and is time consuming. Recent advances in computer vision and deep learning algorithms has helped the automation behavioral annotation. We now have the ability to expand markerless pose estimation techniques to track the position of multiple animals. However, these current methods are still less than optimal in stably tracking the identity of multiple animals in an arena, especially during times of proximity, which in turn is detrimental in the analysis of social behavior. In this thesis, I have identified a pipeline that refines a multi-animal tracking algorithm's output using a combination of an autoencoder architecture and heuristics based outlier detection algorithms to stably maintain animal identity during pose estimation. This pipeline was tested using 35 videos of free-ranging behavior with two black mice and was able to reduce the swapping of the animal identity by approximately 90%. As a pilot test, we used this tracking to align bouts of interaction with neural data recorded from the lateral septum. Upon initial inspection, neural responses in LS seem to be restricted to the first instance of interaction, irrespective of the sex of the animal interacted with, familiarity or sexual experience of the mice. Further directions include optimization of the pipeline and building a stand alone, out-of-the-box ready to go software GUI.
Table of Contents
Introduction……………………………………………………....…………..1 Methods……………………………………………………………….……...3 Results……………………………………………………….……….………7 Discussion…………………………………………………………….…….12 References………………………………………………………………….15
About this Honors Thesis
| School | |
|---|---|
| Department | |
| Degree | |
| Submission | |
| Language |
|
| Research Field | |
| Keyword | |
| Committee Chair / Thesis Advisor | |
| Committee Members |
Primary PDF
| Thumbnail | Title | Date Uploaded | Actions |
|---|---|---|---|
|
|
File download under embargo until 24 May 2026 | 2024-04-29 13:01:47 -0400 | File download under embargo until 24 May 2026 |
Supplemental Files
| Thumbnail | Title | Date Uploaded | Actions |
|---|